Journal of Network and Computer Applications 34 (2011) 765–773

Contents lists available at ScienceDirect

Journal of Network and Computer Applications

journal homepage: www.elsevier.com/locate/jnca

Cross-layer QoS support framework and holistic opportunistic scheduling

for QoS in single carrier WiMAX system
Jinchang Lu n, Maode Ma
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore

a r t i c l e in f o abstract

Article history: Providing Quality of Service (QoS) to different service classes with real-time and non-real-time traffic
Received 7 February 2010 integration is an important issue in WiMAX systems. Opportunistic MAC (OMAC) is a novel view of
Received in revised form communication over spatiotemporally varying wireless links. It combines the features of a cross-layer
5 September 2010
design and an opportunistic scheduling scheme to achieve high utilization while providing QoS to various
Accepted 11 October 2010
Available online 21 October 2010
applications. Channel characteristics, traffic characteristics and queue characteristics are essential factors
in the design of opportunistic scheduling algorithms. In this paper, we propose a cross-layer QoS support
Keywords: scheduling framework and a corresponding opportunistic scheduling algorithm to provide QoS support to
Cross-layer protocol design the heterogeneous traffic in single carrier WiMAX point-to-multipoint (PMP) systems. We model the
QoS
uplink transmission in the single carrier WiMAX system as a multi-class priority TDMA queueing system
Opportunistic scheduling
to analyze the average packet delays of different service classes. Extensive simulation experiments have
WiMAX
been carried out to evaluate the performance of our proposal. The simulation results show that our
proposed solution can improve the performance of the WiMAX PMP system in terms of packet loss rate,
packet delay and system throughput.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction information on the traffic Maximum Latency (ML) constraint,

The WiMAX (IEEE Std 802.16d, 2004) system has received a lot
of attention from the academic and industry sectors in the past few
years as it comes with the ability to provide broadband wireless
access and potential ability to compete with existing wired and
wireless networks. One important issue in the design of WiMAX
systems is to provide QoS to different service classes by using an
efficient scheduling scheme.
Wireless access networks have unique characteristics, which
are the time-varying channel conditions and the multi-user
diversity. The Medium Access Control (MAC) protocol and schedul-
ing algorithms have to be developed specially for this environment
(Liu et al., 2003). It requires a cross-layer MAC protocol design
approach, whereby Channel Specification (Cspec) carrying the
estimated instantaneous channel information can be fed to the
MAC layer from the physical (PHY) layer and Traffic Specification
(Tspec) carrying traffic QoS related information can be fed to the
MAC layer from higher layers such as the network or application
layer. The Cspec feedback includes information on the estimated
instantaneous Signal-to-Interference and Noise Ratio (SINR), sup-
portable data rate R(t), Received Signal Strength Indications (RSSI)
or Bit Error Rate (BER ) of a link. The Tspec feedback includes
n
Corresponding author. Tel.: + 65 91702589.
E-mail address: lu0002ng@ntu.edu.sg (J. Lu).
Maximum Sustained Traffic Rate (MSTR) and the instantaneous
length of queues at a station. The cross-layer nature of OMAC
(Amoakoh et al., 2006) with an opportunistic scheduling scheme
has the potential to revolutionize the design of broadband wireless
access networks from the physical to the networking layer.
Recently, various opportunistic scheduling schemes have been
proposed for wireless communication systems. They can exploit the
time-varying nature of the radio environment to improve the
spectrum efficiency while maintaining a certain level of QoS satisfac-
tion for each connection or user in various wireless networks. The
proposed opportunistic scheduling schemes can be classified into
channel-aware only and channel-aware and queue-aware algorithms
based on their functionality of scheduling algorithms (Hassel, 2006).
Max Carrier-to-Noise Ratio Scheduling (MCS) (Bonald, 2005) is a
typical channel-aware opportunistic scheduling scheme. The MCS
scheme is to allocate resources to users with the best channel
condition to achieve high system throughput. On top of channel
characteristics, traffic characteristics also play an important role in the
design of an opportunistic scheduling algorithm. The Proportional Fair
Scheduling (PFS) schemes (Jalali et al., 2000; Jinri and Niu, 2007; Bang
et al., 2008) attempt to trade-off among the throughput, efficiency and
fairness among users by taking packet length into account (Jinri and
Niu, 2007), or estimating future channel quality (Bang et al., 2008).
Modified Largest Weighted Delay First (M-LWDF) (Andrews et al.,
2000) is a modified version of the PFS scheduler that tries to meet the
QoS requirement in terms of head-of-line packet delay. Traffic-Aided

1084-8045/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jnca.2010.10.005
766 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773
Opportunistic Scheduling (TAOS-1) (Hu et al., 2004) is a heuristic
opportunistic scheduling scheme that unifies file size information and
wireless channel variations in order to reduce the completion time of
file transmission. Exponential Rule Scheduler (EXPRule) (Shakkottai
and Stolyar, 2001) attempts to equalize the weighted delays of all
buffers when their differences become larger in a wireless system.
Comparison of those conventional cross-layer opportunistic schedu-
lers can be found in (Amoakoh and Kim, 2006).
Cross-layer MAC and opportunistic scheduling designs tailored
for WiMAX systems have been proposed (Mai et al., 2007; Lera
et al., 2007; Kwon et al., 2005; Liu et al., 2006; Wan et al., 2007).
TCP-MAC layer cross-layer design has been proposed in the paper
(Mai et al., 2007). It maps network layer 3 and layer 2 function-
alities to provide QoS support in the IEEE 802.16d network. Paper
(Lera et al., 2007) aims at enabling downlink (DL) traffic delivery
with a differentiated service treatment, even in a non-ideal channel
condition. It addresses a channel-aware scheduling algorithm with
a per-flow channel error compensation technique based on the
typical features of a WiMAX system: class-based QoS guarantees
and per-flow resource assignment. The Worst-Case Fair Weighted
Fair Queuing+(WF2Q +) algorithm has been suggested to manage
the flow-level and class-level granularity per-class queues. A cross-
layer adaptive architecture for the IEEE 802.16e OFDMA system has
been proposed in the paper (Kwon et al., 2005). A priority-based
scheduler has been proposed in the paper (Liu et al., 2006). At the
MAC layer, each connection employs the adaptive modulation and
coding (AMC) scheme at the physical (PHY) layer. A Priority
Function (PRF) has been defined for each connection to be admitted
into the system and is updated dynamically depending on the
wireless channel quality, QoS satisfaction and service priority
through a MAC-PHY cross-layer manner. The strategy proposed
in (Wan et al., 2007) has further enhanced the architecture in
(Kwon et al., 2005) by proposing a joint packet scheduling and sub-
channel allocation scheme.
With the help of several additional parameters for the pre-
ference metrics which are the priority of connections or users to be
determined by the opportunistic scheduler for the order of
transmission service, the schemes in (Mai et al., 2007; Lera et al.,
2007; Kwon et al., 2005; Liu et al., 2006; Wan et al., 2007) are shown
to achieve satisfied performance in the given network conditions.
However, the existing scheduling schemes proposed for WiMAX
systems do not consider packet loss rate. The scheduling schemes
have not been applied to packet scheduling in each traffic flow. The
priority coefficient setting for rtPS and nrtPS traffic connections is
rigid while in fact, traffic load and average long-term channel
conditions vary dynamically.
In this paper, we take channel characteristics, queue character-
istics and traffic QoS characteristics of traffic connections into
account, propose a cross-layer QoS support framework and a two-
stage opportunistic scheduling scheme in the single carrier WiMAX
PMP system. The two-stage opportunistic scheduling algorithm is
termed as Holistic Opportunistic Scheduling (HOS) with the
features of channel-awareness, queue-awareness and traffic
QoS-awareness.
Our proposal is able to enhance the proposals in (Liu et al., 2006)
and (Wan et al., 2007) with the following major contributions:
 A cross-layer QoS support framework has been designed to
provide instantaneous Cspec and Tspec feedback to the MAC
layer in the WiMAX system. The detailed signal exchange
protocol is designed to visualize the cross-layer design which
is merely conceptually mentioned in the paper (Liu et al., 2006).
 A novel comprehensive Holistic Opportunistic Scheduling algo-
rithm has been derived with the consideration of not only real-
time physical channel and network traffic conditions but also
traffic QoS requirements and queue status to enhance the
performance of existing opportunistic scheduling algorithms.
The proposed opportunistic scheduling has a unified formula. It
is different from the scheduling scheme in the paper Liu et al.
(2006) which uses different scheduling parameters to schedule
rtPS, nrtPS and BE service classes separately.
 Dynamical priority index is designed for each packet belonging
to the rtPS service class. It can update the priority of rtPS
dynamically according to a linear function based on the channel
condition.
The rest of the paper is organized as follows. In Section 2, we
describe various aspects of the MAC and PHY layers of the WiMAX
system. In Section 3, we present our proposed cross-layer QoS
support framework and the HOS algorithm. In Section 4, we present
the queueing model for the WiMAX system. We model the uplink
transmission of the WiMAX system as a multi-class priority TDMA
queueing system. In Section 5, we show our simulation design and
the simulation results. Finally, in Section 6, we conclude the paper
with a summary.
2. WiMAX PMP system
A WiMAX (IEEE Std 802.16d, 2004) system with PMP topology
includes one Base Station (BS) and M Subscriber Stations (SSs),
where M is the number of SSs. In this study, we focus on the UL
transmission of the WiMAX system in time division duplex (TDD)
mode. The principle of the proposed scheduling algorithm can
be extended to DL. The TDD MAC frame is divided into UL and DL
sub-frames. Service flows which are uniquely identified by a 32-bit
service flow identifier (SFID) may be transmitted in either the
UL or DL sub-frame. Active service flows are associated with
QoS requirement parameters, namely, ActiveQoSParameterSet.
Admitted and active service flows are mapped to a 16-bit connec-
tion identifier (CID), and are controlled and maintained by a
Connection Admission Control (CAC) scheme in Active Connection
List (ACL) at the BS.
The traffic flows at each SS are classified according to the four
types of scheduling services, namely, Unsolicited Grant Service
(UGS), Real-time Polling Service (rtPS), Non-Real-time Polling
Service (nrtPS) and Best Effort (BE) service. The mandatory QoS
service flow parameters for UGS service are Maximum Sustained
Traffic Rate (MSTR), Maximum Latency (ML), Tolerated Jitter and
Request/Transmission Policy. The mandatory QoS service flow
parameters for rtPS service are Minimum Reserved Traffic Rate
(MRTR), MSTR, ML and Request/Transmission Policy. The manda-
tory QoS service flow parameters for nrtPS are MRTR, MSTR, Traffic
Priority and Request/Transmission Policy. The mandatory QoS
service flow parameters for BE are MSTR, Traffic Priority, and
Request/Transmission Policy. As BE connection can tolerate MSTR
and ML QoS requirements, only UGS, rtPS and nrtPS will be
considered in this paper. rtPS and nrtPS can be combined together
under a general polling service umbrella as suggested in the paper
(Ma and Ng, 2006).
The WiMAX system supports 5 different PHY techniques (IEEE
Std 802.16d, 2004), namely, WirelessMAN-SC (Single Carrier) PHY
specification, WirelessMAN-SCa PHY specification, WirelessMAN-
OFDM PHY specification, WirelessMAN-OFDMA PHY specification
and WirelessHUMANTM PHY specification. At the PHY layer, the
WiMAX system takes an adaptive modulation policy which selects
1 out of 3 different modulation schemes. On the UL, QPSK is
mandatory, while 16-QAM and 64-QAM are optional. The DL
supports QPSK and 16-QAM, while 64-QAM is optional. The system
can use various Forward Error Correction (FEC) schemes on the UL
as well as the DL. To fully utilize the flexible and robust PHY layer,
 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773 767

the WiMAX system equips a flexible Radio Link Control scheme, 3. Proposed QoS support framework and scheduling algorithm
which is responsible for transition from one PHY scheme to
another. The RLC is capable of switching between different PHY The performance of MAC protocol and the functionality of
bursts. The system also uses the Receiver Sensitivity as a parameter scheduling algorithms are influenced by the time-varying wireless
together with the SINR thresholds of receivers used by AMC scheme channel in WiMAX systems. In order to enhance the performance of
to select different burst profiles in order to maximize the network MAC protocol with QoS provisioning to the heterogeneous traffic in
throughput and maintain the Bit Error Rate (BER) under a preset WiMAX systems, we propose an efficient cross-layer QoS support
level e.g. BER o ¼1  10  5. framework and a corresponding two-stage opportunistic schedul-
WiMAX transceivers support various transmission modes ing algorithm in the single carrier WiMAX PMP system as shown
with different modulation and coding schemes corresponding to in Fig. 1.
different data transmission rates. For each modulation scheme,
there is one relationship between the theoretical BER and the
3.1. Proposed cross-layer QoS support framework
energy per bit to noise power spectral density ratio (Eb/N0)
(Langton, 2002). By using MATLAB, the diagram of theoretical
BER vs. SINR relationship can be plotted when the PHY channel The proposed cross-layer QoS support framework supports both
bandwidth is set to 25 MHz (IEEE Std 802.16d, 2004) in the single UL and DL transmission. We take the UL transmission as an
carrier WiMAX PMP system. example to explain its functionality as follows:
The range of the received SINR values can be classified into
seven non-overlapping regions for adaptive modulation and coding (1) Considering the impact of air interface on MAC layer protocol, a
index AMC(q) where q¼1,2,y,7 on a target prescribed bit error rate wireless Channel Condition Estimator (CCE) is designed at the
BERo ¼1  10  5 in the WiMAX PMP system. PHY layer at the BS as well as the SSm, where m¼1,2,3,y,M. It
Further complying with the receiver sensitivity RSS require- not only monitors instantaneous channel condition status like
ment specified by the IEEE 802.16d standard (IEEE Std 802.16d, the receiver’s received signal strength RSS(m) and SINR g(m)
2004), a lookup table, labeled Table 1, which is named AMC vs. SINR when the BS receives signal from SSm, but also indicates long
and receiver sensitivity requirement, can be derived when BER is term channel condition by using the statistic parameters
set to 1  10  5. including Max_g(m), Average_g(m) and Min_g(m) over the
execution period. Based on the information of the channel’s
status provided by CCE, the FEC, the Symbol Mapper and the
AMC controller at the BS select an FEC scheme, modulation and
Table 1 coding scheme AMC(m, q), and the data transmission rate for UL
AMC vs. receiver SINR and Min sensitivity requirement.
transmission from SSm to the BS at modulation and coding level
AMC Receiver Receiver Min Modulation/ Channel bit q according to Table 1. By sensing the maximum value of q in
index SINR (dB) sensitivity RSS(q) coding scheme rate R(q) AMC(m, q) at all M SSs in the system, the Symbol Mapper and the
AMC(q) g(q) (dB m) (Mbps) AMC controller at the BS can determine the maximum trans-
mission rate Max_R(t) at time t.
AMC(1) 8.3–11.6  80 BPSK1/2 20
AMC(2) 11.7–13.2  80 QPSK1/2 40 (2) The PHY Symbol Rate Controller at the BS then forms
AMC(3) 13.3–18.9  78 QPSK3/4 60 AMC_Controller_Info{ g(m), RSS(m), AMC(m, q), Max_R(t)} and
AMC(4) 19.0–20.9  73 16-QAM1/2 80 forwards the AMC_Controller_Info to the MAC layer via its PHY
AMC(5) 21.0–28.0  71 16-QAM3/4 120 Service Access Point (SAP). Embedded in UL-Map, the informa-
AMC(6) 28.1–29.1  66 64-QAM2/3 160
AMC(7) 4 ¼29.2  65 64-QAM3/4 180
tion is further broadcasted to all SSs for UL scheduling and
transmission from SSm to the BS.

Fig. 1. Proposed cross-layer QoS support framework for WiMAX.

768 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773
(3) The first stage of the scheduling scheme operates at each SS.
When SSm receives a DL/UL_Map message, its scheduler
extracts g(m), R(m, q) from AMC_Controller_Info. Based on
g(m), R(m, q) and Tspec of each connection, the scheduler at
SSm computes four scheduling parameters and determines the
Scheduling Priority (SP) for every packet in each connection
through our proposed HOS algorithm which is to be described
in detail in Section B. Then, the SPs will be sorted in descending
order. Then, SSm will take the maximum value of the calculated
Scheduling Priority, Max_ SP(m), to represent itself to compete
with other SSs for the order of UL transmission in the next
frame. SSm embeds its Max_SP(m), its total bandwidth request,
BW_request(m) and AMC_Controller_Info in a Polling_respon-
se(m) and sends it to the BS when SSm is polled.
(4) The second stage of scheduling at the BS sorts Max_SP(1),
Max_SP(2),y, Max_SP(M) in descending order and selects SSm
with the highest Max_SP(m) to allocate time slots according to
its BW_request(m) in the next UL sub-frame, excluding the time
slots for UGS. If BW_request(m) is less than the number of
available time slots in the next UL sub-frame, the remaining
time slots will be allocated to SSn which has the second highest
Max_SP(n) according to its BW_request(n). This procedure
iterates until all the available time slots in the next UL
sub-frame are used up. AMC_Controller_Info{ g(m), RSS(m),
AMC(m, q), Max_R(t)} and the result of scheduling, which is
the time slot allocations for SSs in the next UL sub-frame, are
embedded in a DL/UL_Map. The DL/UL_Map is broadcasted by
the BS to all SSs for UL transmission in the next frame.
(5) The scheduler at SSm extracts the information of its allocated
transmission time slots and the AMC_Controller_Info from the
DL/UL_Map and selects packets from the highest priority to
the lowest priority from different connections. SSm transmits
the selected packets to the BS in the allocated time slots at the
data rate determined by AMC(m, q) of the BS receiver.
3.2. Holistic opportunistic scheduling algorithm
In this study, the following design factors have been taken into
consideration when designing the efficient QoS framework and the
corresponding scheduling scheme to provide QoS to the hetero-
geneous traffic in WiMAX systems:
 The existing QoS signaling mechanisms and DL/UL_Map
mechanism in WiMAX should be applied.
 Per-connection scheduling overhead should be minimized.
 The QoS requirement parameters of each connection in the
system should be guaranteed.
 The impact of AMC on the scheduling scheme should be
taken into account as the channel capacity or bandwidth
dynamically fluctuates according to the SINR of PHY in a
wireless time-varying channel.
Considering the design factors mentioned above, we propose
the Holistic Opportunistic Scheduling (HOS) algorithm associated
with the proposed cross-layer QoS support framework. The pro-
posed scheduling scheme takes the information of instantaneous
wireless channel condition and the real-time traffic condition as
well as the traffic QoS specification to set the Scheduling Priority
(SP) for each individual packet for the UL transmission from rtPS or
nrtPS traffic flows. Since the UGS connections have been allocated
with a fixed bandwidth based on their fixed bandwidth require-
ment specified by the framework of the IEEE 802.16d standard, the
proposed scheduling is only applied for the rtPS and nrtPS services.
The SP of each packet is determined by four key scheduling
parameters, namely, Dynamical Priority Index (DPI), Channel
Specification Index (CSI), Normalized Time Delay Satisfaction Index
(NTDSI) and Normalized Predictive Starvation Index (NPSI). In the
scheduling period, starting at time t, SSm (or BS scheduling for DL
traffic) decides the SP of the kth packet from connection i of either
rtPS or nrtPS traffic flows based on the following equation:
SPðm,i,k,tÞ ¼ DPIðm,i,tÞCSIðm,i,tÞ
exp½NTDSIðm,i,k,tÞNPSIðm,i,tÞ ð1Þ
where DPI(m, i, t) is the dynamic priority index of connection i at
SSm. The role of DPI(m, i, t) is to provide different priorities for
different QoS classes dynamically. The priority coefficients are
fixed in the paper (Liu et al., 2006), the priority of rtPS ¼1.0 4the
priority of nrtPS ¼0.8 4the priority of BE ¼0.6. In contrast, we set
priority index, which is termed priority coefficient in the paper (Liu
et al., 2006), of rtPS service class dynamically on a channel-aware
approach. The procedure to set the value of DPI(m, i, t) is described
as follows:
(1) A range for DPI(m, i, t) of [1.0, 0.8] is set for rtPS traffic
connections. SSm firstly assigns 0.8 to DPI(m, i, t) for rtPS traffic
connection i and 0.6 to DPI(m, j, t) for nrtPS traffic j connection.
(2) After the system runs over a period of time, SSm is able to derive
statistical channel condition parameters like Max_g(m),
Average_g(m) and Min_g(m) based on g(m) in AMC_Controller_
Info from the BS. The DPI(m, i, t) for each packet belonging to
rtPS service class can be updated dynamically according to the
linear function:
DPIðm,i,tÞ ¼ 0:8 þ0:2½gðmÞMin_gðmÞ=½Max_gðmÞMin_gðmÞ
ð2Þ
By using (2), DPI(m, i, t) will be set dynamically from 0.8 to 1.0 for
the rtPS service class based on SINR g(m) while it is fixed at 0.6 for
the nrtPS service class. It makes sure that the priority of the rtPS
traffic is always higher than that of the nrtPS traffic. It gives favor
treatment to the rtPS service class to enhance the performance of
QoS for real-time traffic when the channel condition is good.
CSI(m, i, t) in (1) is the channel specification index for connection
i at SSm at time t. It is calculated as follows:
Rðm,i,tÞ
CSIðm,i,tÞ ¼ ð3Þ
Max_RðtÞ
where R(m, i, t) is the transmission rate of connection i at SSm at time
t. It is set according to AMC(m, q) dynamically based on the channel
condition at the PHY layer. Max_R(t) is the maximum of R(m, i, t)
over all M SSs and all connection i at the time instant t. From the
scheduling parameter defined in (3), CSI(m, i, t) is in fact a MCS
scheduling scheme (Bonald, 2005). CSI(m, i, t) is a typical through-
put efficient opportunistic scheduling, which allocates resources to
SSs with the best channel condition. In order to achieve high system
throughput, CSI(m, i, t) plays the role as a channel-aware scheduling
parameter in our proposed HOS.
However, this channel-aware policy monopolizes all resources
to good-channel users that are usually located close to the BS.
Because the SSs are usually distributed across the entire WiMAX
system, this unfair scheduling parameter alone is not beneficial for
anybody but local SSs.
In order to overcome this unfairness caused by the side-effect of
CSI(m, i, t) and to provide delay bound guarantees for individual
connections and delay-sensitive rtPS traffic flows, we introduce a
traffic QoS-aware scheduling parameter, termed as Normalized
Time Delay Satisfaction Index, NTDSI(m, i, k, t), as well as a queue-
aware scheduling parameter, termed as Normalized Predictive
Starvation Index, NPSI(m, i, t) in the HOS scheduling function.
 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773
NTDSI(m, i, k, t) in (1) is the parameter to capture the packet
latency requirement of a packet if applicable. It is defined as
follows:
NTDSIðm,i,k,tÞ ¼ 1TDSIðm,i,k,tÞ=Max½TDSIðm,i,k,tÞ
iAI
Packet timeout is defined as when the waiting time of the packet
in a queue is over its maximum latency ML(m, i, k)or when its
waiting time is over the TCP re-transmission threshold. Time Delay
Satisfaction Index (TDSI) for the kth packet from rtPS connection i at
SSm, TDSI(m, i, k, t), will be set as
( from connection i at SSm at time t is calculated as follows:
MLðm,i,kÞ½tVCðm,i,kÞ if fMLðm,i,kÞ½tVCðm,i,kÞg 4 0
TDSIðm,i,k,tÞ ¼
0, if fMLðm,i,kÞ½tVCðm,i,kÞg r 0
Since a non-real-time packet in nrtPS connections has no
maximum latency requirement, the maximum waiting time for a
non-real-time packet can be set as the TCP re-transmission time-
out value TO at 400 ms for the wireless channel (Song and Mar,
2005). The TDSI(m, i, k, t) for the kth packet from nrtPS connection i
at SSm, TDSI(m, i, k, t), is set as
(
TO½tVCðm,i,kÞ if fTO½tVCðm,i,kÞg4 0
TDSIðm,i,k,tÞ ¼ ð6Þ
0 if fTO½tVCðm,i,kÞgr 0
Each packet is stamped with a virtual clock (VC) (Zhang, 1990).
VC monitors the average transmission rate of statistical data flows
and provides every flow with guaranteed throughput and low queue
delay. The idea behind the VC algorithm is that each packet is
assigned a VirtualTime, which represents the time it would be sent in
a Time Division Multiplexing (TDM) system. The VC(m, i, k) is set for
the kth packet from connection i at SSm according to the algorithm:
VCðm,i,kÞ ¼ Max½VCðm,i,k1Þ,real time þPLðm,i,kÞ=Rðm,i,tÞ
where PL(m, i, k) is the length of the kth packet.[t VC(m, i, k)]
denotes the time spent by the kth packet in its queue.
TDSIðm,i,k,tÞ=Max½TDSIðm,i,k,tÞ is the normalization function so that
iAI
TDSIðm,i,k,tÞ=Max½TDSIðm,i,k,tÞ A ½0,1:
iAI
If
½NTDSIðm,i,k,tÞ ¼ 1TDSIðm,i,k,tÞ=Max½TDSIðm,i,k,tÞ 4 0
iAI
the delay requirement of the kth packet will be satisfied, and the
effect on the scheduling priority will be quantified by exp[  NTDSI
(m, i, k, t)]A[0.368,1]. A smaller value of exp[ NTDSI(m, i, k, t)]
indicates a higher degree of delay satisfaction, which leads to a lower
scheduling priority.
In a wireless network environment, common channel errors due
to multi-path fading, shadowing and attenuation may cause bit
errors and packet loss, which are quite different from the packet
loss caused by congestions in the wired networks. Therefore,
wireless packet loss can mistakenly lead to dramatic performance
degradation. The information of packet loss rate in wireless net-
works not only reflects the condition of the PHY wireless channel
but also serves as an index of effective rate adjustment.
NPSI(m, i, t) in (1) is the normalized predictive starvation index.
It is designed to reflect the packet loss of the system. It is calculated
as follows:
PSIðm,i,tÞ
NPSIðm,i,tÞ ¼
Max½PSIðm,i,tÞ
iAI
where PSI(m, i, t) is the Predictive Starvation Index which indicates
the queue status in terms of queue length and the urgency of a
packet that is to be transmitted successfully:
PSIðm,i,tÞ ¼ the number of packets in queue i at SSm at time t=PSRðm,i,tÞ
769
Here PSR(m, i, t) is the packet success rate with the following
definition: a packet is correctly transmitted from connection i at
SSm at time t, only if it is not dropped from the queue with
ð4Þ probability [1  Pd(m, i, t)] and is correctly transmitted through the
wireless channel with probability [1  PER(m, i, t)]. Hence, we can
obtain the packet loss rate for connection i at SSm at time t, as shown
below:
eðm,i,tÞ ¼ 1½1Pd ðm,i,tÞ½1PERðm,i,tÞ ð10Þ
Therefore, the packet success rate PSR(m, i, t) of the kth packet
PSRðm,i,tÞ ¼ 1eðm,i,tÞ ¼ ½1Pd ðm,i,tÞ½1PERðm,i,tÞ ð11Þ
Let Pd(m, i, t) denote the packet dropping probability, which is
due to its queue being full, the packet delay exceeding its ML or the
waiting time exceeding the TCP re-transmission threshold, upon
the queue for connection i at SSm at time t:
Em,i,t fDg
Pd ðm,i,tÞ ¼ ð12Þ
Em,i,t fAg
Pd(m, i, t) is the ratio of the number of dropped packets Em,i,t{D}over
the number of arrived packets Em,i,t{A} during the sliding window or
scheduling period.
To simplify the AMC design, packet error rate PER(m, i, t) of the
wireless PHY transmission channel for connection i at SSm at time t
can be approximately expressed as
(
1, if gðmÞ r gBS_Min
PERðm,i,tÞ  ð13Þ
an exp½gn gðmÞ if gðmÞ r gBS_Min
ð7Þ gBS_Min is the minimum SINR requirement of the BS receiver.
Parameters an and gn are modulation and coding mode dependent,
which can be determined by the modulation and coding
scheme (Liu et al., 2004).
When a channel condition is good, DPI (m, i, t) will be set at a
high value for the rtPS service class. Thus, a higher system
throughput can be achieved. When R(m, i, t)¼0, the channel is in
deep fading and the capacity is zero, so connection i at SSm should
not be served regardless of its delay performance. The scheduling
scheme can achieve sub-optimality.
3.3. Computional complexity analysis of HOS algorithm
Each SS monitors and computes the SPs of packets from its own
connections. The first stage scheduler at SS only sorts the SPs of
packets in its own queues, the complexity of the algorithm at SS is
O(N log N) where N is the number of packets in the queues.
Similarly, the second stage scheduler at BS only sorts the SS’s
Max_SP(m), where m¼1,2,yM. The scope of the sorting list is
limited to the total number of SSs, the complexity of the algorithm
at BS is O(M log M) where M is the number of SSs. It means that the
overhead of scheduling is divided into two stages and it is
distributed and shared by all SSs. A frame time is sufficient for
every SS to compute four key scheduling parameters for a packet
which could be transmitted in the next frame. The efficiency of the
scheme can be achieved.
ð8Þ
4. Queueing model and delay analysis
To analyze a packet-level QoS in real-time and non-real-time
data transmission, the network performance parameter like aver-
age packet delay needs to be investigated. A queueing analytical
model can be used off-line to obtain the network performance
ð9Þ parameter.
770 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773
In this section, we perform a queueing analytical modeling to
analyze the WiMAX system in order to obtain the average packet
delay for UGS and rtPS service classes.
4.1. WiMAX system queueing model
The UL channel of the single carrier WiMAX PMP system can be
modeled as a multi-class priority TDMA queueing system.
The UL channel can be considered as a multiple-access com-
munication channel shared by M SSs. The UL sub-frame consists of
N consecutive time slots. Each time slot is normalized to be of unit
length t. Let the duration of a time frame be TF, so that TF ¼ Nt. The
TDMA scheme for inter-SS is Max–Min Fair Sharing (MMFS)
scheduling, in which station i is allocated ni time slots per frame,
uniformly distributes in UL sub-frame. The TDMA scheme for inter-
class is the strict priority queue scheme in which higher priority
packets are queued ahead of lower priority ones. Packets of the
same priority class that arrive at different slots are served on a First-
Come-First-Served basis. Packets of the same priority class that
arrive during the same slot are randomly ordered for transmission.
Each station can transmit its packets only during its dedicated time
slots, which is synchronized and governed by the BS via
DL/UL_Map. Packets arriving at each SS belong to one of the four
priority classes: UGS (priority-1, highest priority), rtPS (priority-2),
nrtPS (priority-3) or BE (priority-4, lowest priority). At each station,
packet arrival is a Poisson arrival process so that lj is the average
arrival rate of class j packets, j¼1, 2,y,J. Each packet is transmitted
in different number of time slots according to its length. The
number of time slots to transmit the kth arriving packet of priority
class j is denoted by Bjk (j ¼1, 2,y,J). It is assumed that {Bjk , k Z l} is a
sequence of independent and identically distributed (i.i.d.) random
variables for each class j ¼1, 2,y,J. The transmission requests can
be expressed by integer multiples of a time slot. We assume that the
probability of collision of the request contention in the BE service is
zero in the queueing model. The information of bandwidth alloca-
tion are carried by DL/UP_Map and broadcasted to all SSs. When the
UL channel becomes available, any waiting priority-i packet is
served before any priority-j one, if ioj.
Note that the steady-state average packet delay for each priority
class would be invariant to the intra-slot/intra-frame scheduling
scheme. We focus on the average packet delay of each priority class
regardless of intra-slot/intra-frame scheduling.
4.2. Delay analysis
The delay of a packet is defined as the total time spent by the
packet to get through UL channel. Denoted byDjk , the delay of the
kth packet of priority class j, we have
Djk ¼ Wkj þ sjk þ Pkj þFL
where Wkj denotes the packet waiting time, sjk represents the total
time to transmit a priority—j kth packet. FL denotes the packet
frame latency. Pkj denotes the packet propagation delay. The
average packet delay of j class is Dj , given by (Moraes and Rubin,
1984)
 h i  
N1 N
D j ¼ Wj,1 þ E Bjk  t þ FL þ P j ð15Þ
N ni
where
j k
PJ N
i¼1
li bi,2 þ ð1ZJ Þ ni t percentages of rtPS and nrtPS traffic loads from the eight SSs to
Wj,1 ¼
2ð1Zj Þð1Zj1 Þ
 
N=ni represents the biggest integer not larger than nNi .
bj,2 is the second moment of service time, and is given by
Z 1
bj,2 ¼ x2 dBj ðxÞ
0
rj ¼ lj bj,1
j
X
Zj ¼ ri , j ¼ 1,2,. . .,J ð17Þ
i¼1
For UGS traffic, we set J ¼1. And we can set J¼2, J ¼3, J ¼4 for
rtPS traffic, nrtPS and BE traffic, respectively.
5. Performance evaluation
Following the signaling mechanism specified by the IEEE
802.16d standard, we developed our own WiMAX simulation
model by using C language. We have carried out extensive
simulation experiments to evaluate the performance of the pro-
posed cross-layer QoS support framework and the HOS algorithm.
We compare the performances of Exponential Rule Scheduler
(EXPRule) in Shakkottai and Stolyar (2001) and Priority Function
Scheduler (PRFS) in Liu et al. (2006) with that of our scheduling
scheme for rtPS and nrtPS connections. Average packet delays of
service classes obtained by quantitative analysis of the queueing
model have been compared with the simulation results. The
simulation results show that our proposed scheduling scheme
can not only effectively support rtPS traffic to make real-time
packets meet their delay bounds but also serve nrtPS traffic with
reasonable performance.
5.1. Simulation design
For the wireless fading channel in the simulation model, the
channel quality can be captured by the parameter, g. Since the
transmission channel varies from frame to frame, we employed
the general Nakagami-m model to describe g statistically
(Nakagami, 1960). The received SINR, g per frame is thus a random
variable with a Gamma probability density function:

mm gm1 mg
pg ðgÞ ¼ m exp  ð18Þ
g GðmÞ g
R1
where g is the average received SINR, GðmÞ : ¼ 0 t m1 et dt is
the Gamma function and m is the Nakagami fading parameter
(m Z1/2). We set m ¼1 in our simulation model. The channel is
assumed to be quasi-static so that the channel gains are constant
over a frame. However, they are allowed to vary from frame to
frame. The average received SINR, gcan be expressed as
g ¼ Pt Li PN LP ð19Þ
where Pt s the transmitter output power, Li is the implementation
ð14Þ loss due to hardware connecting cables, antenna patterns, etc.,
PN is the receiver noise power, which is related to the hardware
noise figure and bandwidth, and LP is the path loss due to radio
propagation.
The network has been designed with one BS and eight SSs. The
positions of the SSs are assumed to be independent and distributed
randomly. The traffic parameters are selected from the supported
range of values which fully cover the required values for multi-
media services defined in Recommendation ITU-R M.1225 (ITU
RADIO, 2007). In the simulation experiments, we only consider the
effect of the UL scheduling algorithm on the integrated traffic
consisting of a fixed percentage of UGS traffic load, different
ð16Þ the BS. We set UGS at 1.37 Mbps. The traffic load for rtPS and nrtPS
is 50% each of the remaining traffic. The inter-arrival time of UGS
packets is constant, at 10.1 ms. Packet length is fixed to 200 bytes.
 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773 771

The deadline is set to 50 ms. Both rtPS and nrtPS connections are reason is that the PRFS scheme has only considered the delay
burst traffic flows. Packet arrivals follow the Poisson distribution satisfaction indicator for the rtPS traffic and the EXPRule scheme
with packet inter-arrival time exponentially distributed. Packet has only considered starvation index for the rtPS traffic. However,
length is exponentially distributed with a mean packet length of our scheduler has integrated both Normalized Time Delay Satisfac-
500 bytes. The mean deadline of rtPS is 80 ms and it is exponen- tion Index NTDSI(m, i, k, t) and Normalized Predictive Starvation
tially distributed. There is no deadline for nrtPS connections. The Index NPSI(m, i, t) into the computation of scheduling decision for
queue size is big enough so that packets would not be dropped due
to the queue being full. We consider packet loss due to its waiting
time exceeding its maximum latency or packet destroyed because 30
of interference and fading of the wireless channel in our simulation.
The parameters for simulation experiments are shown in PRFS
Table 2. 25
EXPRule
5.2. Numerical calculation of average packet delay 20 HOS

rtPS Loss Rate %

According to the system configuration, we can get ni ¼250/
8¼31 slots/frame as all SSs are identical. Note that B1 ¼ 4, 15
B2 ¼ 10,B3 ¼ 10 according to the average packet length of the 3
different service classes. By using (15), the numerical results of the
average packet delay of UGS, rtPS and nrtPS classes can be 10
calculated as shown in Table 3. It has been evaluated and compared
with the simulation results of the EXPRule, PRFS and HOS schedul- 5
ing algorithms.

5.3. Experimental results 0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
In the simulation experiments, it is assumed that the system Total Traffic Load
operates in TDD mode. As shown in Figs. 2–6, we evaluate the Fig. 2. Loss rate vs. traffic load for rtPS traffic.
performance of the three scheduling schemes in terms of packet
loss rate, average packet delay and UL throughput of the rtPS
connections, entire UL throughput and UL throughput of the nrtPS
300
connections with the total traffic load, which consists of UGS, rtPS
and nrtPS service classes, changes from 10% to 90% of the total UL
bandwidth accordingly. PRFS
250
Fig. 2 shows the relationship between the packet loss rate of the EXPRule
rtPS traffic and the various traffic loads. It is clear that the packet
HOS
loss rate of the rtPS traffic of our proposal is lower than that of the 200
rtPS Delay (ms)

PRFS and EXPRule schemes when the traffic intensity increases. The Queueing Model

150
Table 2
Simulation parameters based on IEEE 802.16d SC system.

Parameters Value 100

Channel bandwidth 25 MHz

Frame length 10 ms 50
UL sub-frame length 5 ms
Duration of time slot 20 ms
Average received SINR, g 12 dB 0
Adaptive modulation scheme BPSK, QPSK, M-QAM
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Target BER, Pber r 10  5
Wireless channel model Nakagami-m fading channel (m ¼1) Total Traffic Load
Receiver sensitivity, RSS  65 dB m
Fig. 3. rtPS delay vs. traffic load.

Table 3
Numerical calculation of Mean delay.

Traffic Load UGS (Mbps) rtPS (Mbps) nrtPS (Mbps) k1 (mean) Packet/s k2,3 (mean) Packet/s UGS Delay(ms) rtPS Delay(ms) nrtPS Delay(ms)

0.1 1.37 0.315 0.315 107.03 9.84 6.14 7.91 8.72

0.2 1.37 1.315 1.315 107.03 41.09 6.14 10.48 13.98
0.3 1.37 2.315 2.315 107.03 72.34 6.14 13.09 19.44
0.4 1.37 3.315 3.315 107.03 103.59 6.14 15.73 25.12
0.5 1.37 4.315 4.315 107.03 134.84 6.14 18.41 31.02
0.6 1.37 5.315 5.315 107.03 166.09 6.14 21.11 37.15
0.7 1.37 6.315 6.315 107.03 197.34 6.14 23.85 43.53
0.8 1.37 7.315 7.315 107.03 228.59 6.14 26.63 50.17
0.9 1.37 8.315 8.315 107.03 259.84 6.14 29.45 57.07
772 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773

0.5 0.5

PRFS PRFS
0.4 EXPRule 0.4 EXPRule
HOS
HOS
rtPS UL Throughput

nrtPS UL Throughput
0.3 0.3

0.2 0.2

0.1
0.1

0.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Total Traffic Load
Total Traffic Load
Fig. 4. rtPS UL throughput vs. traffic load.
Fig. 6. nrtPS UL throughput vs. traffic load.

0.9
It is also observed that the average packet delays of the queueing
0.8 model and HOS are very close to each other. The simulation results
PRFS have been validated by the queueing model.
0.7 Fig. 4 shows the relationship between the UL throughput of the
EXPRule rtPS connections with traffic load for the three scheduling schemes.
System UL Throughput

0.6 It can be observed that our proposal can achieve a higher UL

HOS
0.5
0.4
0.3
0.2
0.1
0.0
0.1 0.2 0.3
Fig. 5. System UL throughput vs. traffic load.
the transmission priority. Therefore, it can achieve a much lower
packet loss rate for the rtPS traffic connections.
Fig. 3 shows the average packet delays of the rtPS connections of
the three scheduling schemes and the numerical calculation results
of the queueing model. It is clear that the EXPRule scheduler
degrades dramatically but PRFS and our scheduler do not. The
reason is that the delay requirement of a packet from a rtPS
connection has been included to determine its SP by exp[-NTDSI(m,
i, k, t)]A [0.368,1]. If ML(m, i, k) (t  VC(m, i, k)r0, a lower value of
NTDSI(m, i, k, t) can be set, leading to a higher SP. The expression
exp[  NTDSI(m, i, k, t)] makes the scheduling priority assignment
much more sensitive to the delay bound of a packet from the rtPS
traffic.
It is observed that the average packet delays of the queueing
model and that of the three schemes are very close to each other
when the traffic load is light, for example, less than 0.6. This is based
on the fact that the system can transmit packets in time when the
traffic load is light.
throughput of the rtPS service class.
Based on (2), the DPI assignment of our proposal assigns rtPS
traffic with a higher DPI when channel condition is good to provide
more transmission opportunities for the rtPS traffic bursts. Further-
more, the delay requirement of a packet from a rtPS connection has
been included to determine its SP by exp[ NTDSI(m, i, k, t)]A
[0.368,1]. If ML(m, i, k) (t  VC(m, i, k) r 0, a lower value of
NTDSI(m, i, k, t) can be set, leading to a higher SP for a rtPS packet.
Fig. 5 shows the system UL throughput with different traffic
loads. It is observed that our proposal can achieve a higher UL
throughput. It is clear that EXPRule scheduler and PRFS degrade
dramatically but our scheduler does not when the traffic load is
0.4 0.5 0.6 0.7 0.8 0.9 more than 0.7. The connection-based scheduler could waste
Total Traffic Load bandwidth, if the BW_ request of a scheduled connection is less
than the available bandwidth in a frame. Since the HOS scheduler is
a packet-based scheduler without any waste of available band-
width, efficient bandwidth utilization can be achieved by our
proposal. Furthermore, the EXPRule scheme has not considered
the ML requirement of the rtPS traffic, the packets from rtPS
connections drop drastically when traffic load is heavy, e.g. more
than 0.7. This leads to a lower UL throughput by the EXPRule
scheme.
Fig. 6 shows the UL throughput of nrtPS connections with
different traffic loads. It is clear that our solution can achieve a
moderately higher nrtPS UL throughput compared to PRFS and
EXPRule. The reason is that we have included the predictive
starvation index NPSI (m, i, t) in the opportunistic scheduling
design. When the DPI (m, i, t), NTDSI(m, i, k, t) and CSI (m, i, t) are the
same for all the connections, NPSI (m, i, t) becomes the major factor
in determining the Scheduling Priority. When the queue length of
one of the nrtPS connections increases, it leads to a higher value of
NPSI (m, i, t), so the starvation problem of nrtPS connections can be
mitigated. The queue length of each connection will be restricted to
a certain level to make the system stable.
The EXPRule scheduling scheme is a connection-based scheduler.
The bandwidth can be wasted, if the BW_ request of a scheduled
 J. Lu, M. Ma / Journal of Network and Computer Applications 34 (2011) 765–773
connection is less than the available bandwidth in a frame. The
EXPRule scheduling has considered the transmission rate optimal for
connections. However, this factor monopolizes all resources to good-
channel users that are usually located close to the BS. Because the SSs
are usually distributed across the entire WiMAX system, this unfair
scheduling parameter alone is not beneficial for anybody but local
SSs. This is why the nrtPS UL throughput by EXPRule is lower than
that of PRFS and HOS schemes.
6. Conclusion
In this paper, we have proposed a cross-layer QoS support
framework to enhance QoS provisioning specified by the IEEE
802.16d standard in the single carrier WiMAX PMP system.
Associated with the framework, the proposed HOS algorithm,
which has the features of channel-awareness, queue-awareness
and traffic QoS-awareness, determines the dynamic SP of each
packet by its four key scheduling parameters, namely the Dyna-
mical Priority Index, the Channel Specification Index, the Normal-
ized Time Delay Satisfaction Index and the Normalized Predictive
Starvation Index. Our proposal can effectively and efficiently
schedule and manage the transmission of the integrated traffic
consisting of UGS, rtPS and nrtPS connections. The simulation
results show that the proposed solution can improve QoS sig-
nificantly for rtPS traffic connections while achieving a high UL
throughput.
References
Andrews A, Kumaran K, Ramanan K, Stolyar A, Vijayakumar R,Whiting P. CDMA data
QoS scheduling on the forward link with variable channel conditions. Bell Labs
technical report preprint, April 2000.
Amoakoh Gyasi, Seong Agyei, Kim Lyun. Comparison of opportunistic scheduling
policies in time-slotted AMC wireless networks. In: Proceedings of the IEEE first
international symposium on wireless pervasive computing, 2006, 6pp.
Bang HJ, Ekman T, Gesbert D. Channel predictive proportional fair scheduling. IEEE
Transactions on Wireless Communications February 2008;7(2):482–87.
Bonald T. Flow-level performance analysis of some opportunistic scheduling
algorithms. European Transactions on Telecommunications 2005;16:65–75.
Hassel Vegard. Tutorial: opportunistic multiuser scheduling for wireless networks.
Department of Electronics and Telecommunications, Norwegian University of
Science and Technology (NTNU), June 2006. Available from: /www.unik.no/
personer/porten/MoPSAR/MOPSARTutorial_Hassel.pdfS.
Hu M, Zhang J, Sadowsky J. Traffic aided opportunistic scheduling for wireless
networks: algorithms and performance bounds. Computer Networks: The
International Journal of Computer and Telecommunications Networking
2004;46(4):505–18.
773
IEEE Std 802.16d-2004. IEEE standard for local and metropolitan area
networks—Part 16: air interface for fixed broadband access systems, October
2004.
ITU RADIO COMMUNICATION STUDY GROUPS. Document 8F/1359-E, August 2007.
Available from: /http://members.wimaxforum.org/documents/WiMAX_IMT_2000/
itu/8F_1359.docS.
Jalali A, Padovani R, Pankaj R. Data throughput of CDMAHDR: a high efficiency-high
data rate personal communication wireless system. In: Proceedings of the IEEE
VTC 2000, May 2000. p. 1854–8.
Jinri Huang, Niu Zhisheng. A cross-layer proportional fair scheduling algorithm
with packet length constraint in multiuser OFDM networks. In: Proceedings of
the IEEE international conference on global telecommunications conference,
GLOBECOM ’07, November 2007. p. 3489–93.
Kwon T, Lee H, Choi S, Kim J, Cho D-H, Cho S, et al. Design and implementation of a
simulator based on a cross-layer protocol between MAC and PHY layers in a
WiBro compatible IEEE802.16e OFDMA system. IEEE Communication Magazine
December 2005;43(12):136–46.
Langton Charan, Intuitive Guide to Principles of Communications, 2002. Available
from: /www.complextoreal.com/chapters/fm.pdfS.
Lera A, Molinaro A, Pizzi S. Channel-aware scheduling for QoS and fairness
provisioning in IEEE 802.16/WiMAX broadband wireless access systems. IEEE
Network 2007;21(5):34–41.
Liu Q, Zhou S, Giannakis GB. Cross-layer modeling of adaptive wireless links for QoS
support in multimedia networks. In: Proceedings of the IEEE first international
conference on quality of service in heterogeneous wired/wireless networks,
2004. p. 68–75.
Liu Q, Wang X, Giannakis GB. A cross-layer scheduling algorithm with QoS support in
wireless networks. IEEE Transactions on Vehicular Technology 2006;55(3):
839–47.
Liu Xin, Chong EKP, Shroff NB. Optimal opportunistic scheduling in wireless
networks. Proceedings of the IEEE 58th vehicular technology conference
2003e;3:1417–421.
Ma M, Ng BC. Supporting differentiated services in wireless access networks. In:
Proceedings of the IEEE international conference on communication systems,
2006. p. 1–5.
Moraes L, Rubin I. Message delays for a TDMA scheme under a nonpreemptive
priority discipline. IEEE Transactions on Communications 1984;32(5):583–8.
Nakagami M. The m-distribution—a general formula of intensity distribution of
rapid fading. In: Hoffman WG, editor. Statistical methods in radio wave
propagation: proceedings of a symposium held at the University of California.
Pergamon Press; 1960. p. 3–36.
Shakkottai S, Stolyar A. Scheduling algorithms for a mixture of real-time and non-
real-time data in HDR. In: 17th international teletraffic congress (ITC-17),
September 2001.
Song G, Li Y. Cross-layer optimization for OFDM wireless networks—Part II:
algorithm development. IEEE Transactions on Wireless Communication March
2005;4(2):625–34.
Wan Lihua, Wenchao Ma, Zihua Guo. A cross-layer packet scheduling and
subchannel allocation scheme in 802.16e OFDMA system. In: Wireless com-
munications and networking conference, 2007. p. 1865–70.
Mai Yi-Ting Yang, Chun-Chuan, Lin Yu-Hsuan. Cross-layer QoS framework in the
IEEE 802.16 network. In: The ninth international conference on advanced
communication technology, vol. 3, February 2007. p. 2090–95.
Zhang L. Virtual clock: a new traffic control algorithm for packet switching networks.
In: Proceedings of the ACM SIGCOMM’90, Philadelphia, PA, vol. 20(4), September
1990. p. 19–29.